Separations¶
When linear manifold is formed, a distance from every point of dataset to the manifold is calculated, and a histograms of point distances to each trial manifold are computed. If the resulting histogram contains multiple modes then the mode near zero is isolated in histogram [1]. The isolated part of histogram is used to determine a separation criteria, and the data points are partitioned from the rest of the dataset on the basis of such separation.
The separation properties defined in Separation
instance, which is defined as follows:
type Separation depth::Float64 # Separation depth (depth between separated histogram modes) discriminability::Float64 # Separation discriminability (width between separated histogram modes) threshold::Float64 # Distance threshold value globalmin::Int # Global minimum as histogram bin index hist_range::Vector{Float64} # Histogram ranges hist_count::Vector{UInt32} # Histogram counts bin_index::Vector{UInt32} # Point to bin assignments end
Separation criteria and distance threshold value can be accessed through following functions:
-
criteria
(S)¶ Returns separation criteria value which is product of depth and discriminability.
-
threshold
(S)¶ Returns distance threshold value for separation calculated on histogram of distances. It is used to determine which points belong to formed cluster.
References
[1] |
|